An approximate positions-of-people determination means determines, for each partial area, approximate positions of people on an image on the basis of information about the partial area. A states-of-people estimation means, in addition to predicting the current states of particles from the states of particles indicating past states of people and stored in a storage means, adds new particles and evaluates the likelihood of the states of predicted particles and the added particles on the basis of an observation model generated for the observed people to update the weights of the particles, performs particle re-sampling at a probability proportional to the weights of those particles, and outputs the states of the obtained particles as a flow of people.
Legal claims defining the scope of protection, as filed with the USPTO.
1. A people flow estimation device comprising: a memory storing a computer program; and at least one processor executing the computer program to perform: estimating, for each partial area of an image, a number of persons within the partial area, based on a partial area image to be specified in the partial area; correcting the number of persons within the partial area to an integer, determining, for the each partial area, an approximate position of a person by the corrected number of persons on the image, based on information relating to the partial area, calculating a number-of-people correction coefficient, based on the number of persons before correction and the number of persons after correction, and determining that the person is observed at the approximate position; generating an observation model relating to an observation value representing the person, based on the approximate position of the observed person, for the each observed person, and information relating to the partial area; and predicting a current state of one or more particles from a past state of the particles representing a past state of the person stored in a storage, adding new particles and updating a weight of the particles by evaluating probability of the predicted state of the particles and a state of the added particles, based on an observation model generated for each observed person and the number-of-people correction coefficient, storing, in the storage, the state of the particles acquired by resampling the particles with a probability proportional to the updated weight, and outputting the acquired state of the particles as information on a flow of people.
2. The people flow estimation device according to claim 1 , wherein the at least one processor performs determining, as the approximate position of the person, a random position within the partial area, a center position of the partial area, or a random position in a person's area within the partial area.
3. The people flow estimation device according to claim 1 , wherein the at least one processor performs calculating the number-of-people correction coefficient by dividing the number of persons before correction by the number of persons after correction.
4. The people flow estimation device according to claim 1 , wherein the at least one processor performs generating, for each observed person, the observation model in which at least one of a position of the person on the image and a size of the person on the image is set as the observation value, based on the approximate position of the observed person, and information relating to the partial area.
5. The people flow estimation device according to claim 1 , the at least one processor performs: acquiring motion information from the image, and generating, for each observed person, the observation model in which at least one or both of a speed and a moving direction of the person on the image is set as the observation value, based on the approximate position of the observed person, the motion information associated with the approximate position of the person, and information relating to the partial area.
6. The people flow estimation device according to claim 1 , the at least one processor performs: acquiring motion information from the image, and generating, as a histogram on at least one or both of a speed and a direction, for each observed person, the observation model in which at least one or both of a speed and a moving direction of the person on the image is set as the observation value, based on the motion information within the partial area associated with the observed person.
7. The people flow estimation device according to claim 1 , wherein the at least one processor performs determining a number of particles to be added as the new particles, for each partial area, by calculating a product of the number of persons within the partial area, a number of particles per person, and an appearance probability of the person, and adds the new particles by the determined number of particles at a position based on the associated partial area.
8. The people flow estimation device according to claim 1 , the at least one processor performs extracting one or more particles passing a predetermined crossed line in each passing direction, based on a current state of the particles output as the flow of people, a past position of the particles estimated from a state of the particles, a past state of the particles output as the flow of people at a previous time, and a current position of the particles predicted from a state of the particles, and calculating a number of persons passing the crossed line in each passing direction by using the weight of the extracted particles.
9. The people flow estimation device according to claim 1 , the at least one processor performs when the particles are subjected to clustering, and a closed space for surrounding the particles belonging to a same cluster is set, or when the closed space is set from an outside, calculating the number of persons within the closed space from the weight of the particles within the closed space, calculating a representative velocity from a moving velocity of the particles within the closed space, and displaying, on a display device, the closed space, the number of persons, an arrow representing the representative velocity, and a numerical value or a phrase associated with the representative velocity.
10. A people flow estimation method comprising: estimating, for each partial area of an image, a number of persons within the partial area, based on a partial area image to be specified in the partial area; correcting, for the each partial area, the number of persons within the partial area to an integer, determining an approximate position of a person by the corrected number of persons on the image, based on information relating to the partial area, calculating a number-of-people correction coefficient, based on the number of persons before correction and the number of persons after correction, and determining that the person is observed at the approximate position; generating, for each observed person, an observation model relating to an observation value representing the person, based on the approximate position of the observed person and information relating to the partial area; and predicting a current state of the particles from a state of the particles representing a past state of the person stored in a storage, adding new particles and updating a weight of the particles by evaluating probability of the predicted state of the particles and a state of the added particles, based on an observation model generated for each observed person and the number-of-people correction coefficient, storing, in the storage, a state of the particles acquired by resampling the particles with a probability proportional to the updated weight, and outputting the acquired state of the particles as information on a flow of people.
11. The people flow estimation method according to claim 10 , further comprising: determining, as the approximate position of the person, a random position within the partial area, a center position of the partial area, or a random position in a person's area within the partial area.
12. The people flow estimation method according to claim 10 , further comprising: calculating the number-of-people correction coefficient by dividing the number of persons before correction by the number of persons after correction.
13. The people flow estimation method according to claim 10 , further comprising: generating, for each observed person, the observation model in which at least one of a position of the person on the image and a size of the person on the image is set as the observation value, based on the approximate position of the observed person and information relating to the partial area.
14. The people flow estimation method according to claim 10 , further comprising: acquiring motion information from the image; and generating, for each observed person, the observation model in which at least one or both of a speed and a moving direction of the observed person on the image is set as the observation value, based on the approximate position of the observed person, the motion information associated with the approximate position, and information relating to the partial area.
15. The people flow estimation method according to claim 10 , further comprising: acquiring motion information from the image; and generating, for each observed person, the observation model in which at least one or both of a speed and a moving direction of the observed person on the image is set as the observation value, as a histogram on at least one or both of the speed and the direction, based on the motion information within the partial area associated with the observed person.
16. The people flow estimation method according to claim 10 , further comprising determining a number of particles to be added as the new particles, for each partial area, by calculating a product of the number of persons within the partial area, a number of particles per person, and an appearance probability of the person, and adding particles by the number of particles at a position based on the partial area associated with the person.
17. A non-transitory recording medium having a people flow estimation program recorded therein, the people flow estimation program causing a computer to execute: number-of-people estimation processing of estimating, for each partial area of an image, a number of persons within the partial area, based on a partial area image to be specified in the partial area; approximate one's position determining processing of correcting the number of persons within the partial area to an integer, for the each partial area, determining an approximate position of the person by the corrected number of persons on the image, based on information relating to the partial area, calculating a number-of-people correction coefficient, based on the number of persons before correction and the number of persons after correction, and determining that the person is observed at the approximate position; observation model generation processing of generating, for the each observed person, an observation model relating to an observation value representing the person, based on the approximate position of the observed person and information relating to the partial area; and states-of-people estimation processing of predicting a current state of one or more particles from a state of the particles representing a past state of the person stored in a storage, adding new particles and updating a weight of the particles, by evaluating probability of the predicted state of the particles and a state of the added particles, based on the observation model generated for each observed person and the number-of-people correction coefficient, storing, in the storage, a state of the particles acquired by resampling the particles with a probability proportional to the updated weight, and outputting the acquired state of the particles as information on a flow of people.
Cooperative Patent Classification codes for this invention. Click any code to explore related patents in that topic.
September 11, 2017
October 20, 2020
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.